NTIRE 2026 年短时 UGC 视频野外恢复挑战：数据集、方法与结果
计算机视觉与模式识别
2026-04-14 v1
摘要
本文概述了 NTIRE 2026 年短时用户生成内容（UGC）视频野外恢复挑战。该挑战采用由中国科学技术大学和快手科技贡献的全新短时 UGC（S-UGC）视频恢复基准数据集 KwaiVIR，包含野外环境下的人工失真视频和真实短时 UGC 视频。本次公开数据包括 200 个人工失真训练视频、48 个野外训练视频、11 个验证视频以及 20 个测试视频。该挑战的主要目标是为在复杂真实环境下恢复短时 UGC 视频建立强大且实用的基准，尤其聚焦于新兴的基于生成模型的 S-UGC 视频恢复范式。本次挑战设有两个赛道：（i）主赛道为主观赛道，评估基于用户研究；（ii）第二个赛道为客观赛道。这两个赛道使我们能够全面评估恢复质量。总计有 95 支团队报名参赛，12 支团队提交了有效的最终解决方案和测试阶段的事实表。提交的方法在 KwaiVIR 基准上取得了强劲表现，显示了在野外环境下恢复短时 UGC 视频方面取得了令人鼓舞的进展。
引用
@article{arxiv.2604.10551,
title = {NTIRE 2026 Challenge on Short-form UGC Video Restoration in the Wild with Generative Models: Datasets, Methods and Results},
author = {Xin Li and Jiachao Gong and Xijun Wang and Shiyao Xiong and Bingchen Li and Suhang Yao and Chao Zhou and Zhibo Chen and Radu Timofte and Yuxiang Chen and Shibo Yin and Yilian Zhong and Yushun Fang and Xilei Zhu and Yahui Wang and Chen Lu and Meisong Zheng and Xiaoxu Chen and Jing Yang and Zhaokun Hu and Jiahui Liu and Ying Chen and Haoran Bai and Sibin Deng and Shengxi Li and Mai Xu and Junyang Chen and Hao Chen and Xinzhe Zhu and Fengkai Zhang and Long Sun and Yixing Yang and Xindong Zhang and Jiangxin Dong and Jinshan Pan and Jiyuan Zhang and Shuai Liu and Yibin Huang and Xiaotao Wang and Lei Lei and Zhirui Liu and Shinan Chen and Shang-Quan Sun and Wenqi Ren and Jingyi Xu and Zihong Chen and Zhuoya Zou and Xiuhao Qiu and Jingyu Ma and Huiyuan Fu and Kun Liu and Huadong Ma and Dehao Feng and Zhijie Ma and Boqi Zhang and Jiawei Shi and Hao Kang and Yixin Yang and Yeying Jin and Xu Cheng and Yuxuan Jiang and Chengxi Zeng and Tianhao Peng and Fan Zhang and David Bull and Yanan Xing and Jiachen Tu and Guoyi Xu and Yaoxin Jiang and Jiajia Liu and Yaokun Shi and Wei Zhou and Linfeng Li and Hang Song and Qi Xu and Kun Yuan and Yizhen Shao and Yulin Ren},
journal= {arXiv preprint arXiv:2604.10551},
year = {2026}
}
备注
Accepted by CVPR 2026 workshop; NTIRE 2026